Marking PlutusTx.Builtins.Internal -O is overkill

[treeowl]

Currently,

-- This ensures that we don't put *anything* about these functions into the interface
-- file, otherwise GHC can be clever about the ones that are always error, even though
-- they're NOINLINE!
{-# OPTIONS_GHC -O0 #-}

Now I don't know what the problem is with GHC being clever like that (that should be documented!), but this is overkill regardless. As far as I can tell, the only thing that's always error is error :: BuiltinUnit -> a. Everything else in the module should ideally be analyzed properly. I think the right thing is to just tell GHC exactly what we mean.

error :: BuiltinUnit -> a
error (BuiltinUnit ()) = GHC.Exts.lazy (mustBeReplaced "error")

Note: in combination with #4193, we'll then want to use lazy in the BuiltinByteString functions so GHC sees those as strict in their BuiltinByteString arguments but not in the underlying ByteStrings.

[michaelpj]

Yes, it's a blunt hammer, but it works. What we really want is ghc-proposals/ghc-proposals#415 (for essentially the same reason as the Clash folks want it).

[effectfully]

Yes, it's a blunt hammer, but it works.

Given this, I'm marking this issue with "Low priority".

[effectfully]

I think it's fine to have -O0 in there, so rather than spending time investigating this issue, I'm going to close it instead. Feel free to reopen if you have a strong feeling about it.